Glomerular filtration rate-estimating equations for patients with advanced chronic kidney disease

Renal function is often estimated using one of several glomerular filtration rate (GFR) estimating equations. However, there is no consensus which estimating equation performs best in patients with advanced renal failure. We compared the performance of five different estimated GFR (eGFR) equations [Cockcroft-Gault, Modification of Diet in Renal Disease (MDRD), Chronic Kidney Disease (CKD) Epidemiology collaboration (CKD-EPI) and Mayo Clinic and Lund-Malm] with measured GFR (plasma iohexol clearance) in 2098 referred CKD patients with mGFR 30 mL/min/1.73 m(2). There were 398 patients with an mGFR 10 mL/min/1.73 m(2), 1974 with a measured GFR (mGFR) 1120 mL/min/1.73 m(2) and 749 patients with mGFR 2130 mL/min/1.73 m(2). Across the entire range, the median bias of eGFR was lowest for the Lund-Malm equation (0.7 mL/min/1.73 m(2)), followed by the CKD-EPI (1.2 mL/min/1.73 m(2)), the MDRD (1.6 mL/min/1.73 m(2)), Mayo Clinic equation (1.7 mL/min/1.73 m(2)) and Cockcroft-Gault equation (4.6 mL/min/1.73 m(2)). The best accuracy within 30 of mGFR was also for Lund-Malm (76), while it was similar for CKD-EPI, MDRD and Mayo (6567). The Cockcroft-Gault had the worst accuracy of only 54.The median bias was stable across mGFR categories, while the accuracy within 30 of mGFR became worse with decreasing mGFR. All equations performed best among patients with hereditary kidney diseases and tubulointerstitial disease. Accuracy was generally worse for patients 65 years of age and for those with diabetic nephropathy. In patients with advanced renal failure, the GFR-estimating equations show reasonably good performance on the population level. On the individual patient level, they are inaccurate, especially in elderly patients and those with diabetic nephropathy.